Alcove MCP server provides focused, local documentation for agents
alcove, by Epicsagas, is an MCP server that stores private project documentation to prevent context bloat in AI coding agents. It delivers on-demand, searchable context to MCP-compatible agents using hybrid BM25 plus vector retrieval, enabling agents to request narrow documentation snippets instead of entire files. Built in Rust with hybrid search, policy enforcement, semantic linting, zero-config project switching, CJK support, and local-first architecture, it targets developers, technical writers, and engineering teams who need tighter context control and improved agent relevance.
What tasks you can actually use it for
As an MCP server, the tool functions as a private documentation library that agents query at generation time. Use cases include supplying architecture decisions, API conventions, code examples, and targeted doc excerpts to coding assistants so prompts avoid excess material. The interface serves those requests on demand, letting agents pull specific context fragments rather than embedding large files in each prompt, which directly addresses prompt token limits in agent workflows.
How precise its retrieval is for developer queries
Alcove uses a hybrid search approach combining BM25 keyword scoring with vector similarity to improve retrieval precision when agents ask for context. That design makes it more likely agents receive tightly relevant passages compared to simple keyword-only lookups, and the system includes explicit handling for CJK text to avoid common multi-language retrieval errors. Retrieval quality still tracks the clarity and structure of the stored documentation.
What inputs and environments it accepts
The tool speaks the Model Context Protocol, so any MCP-compatible host can request context; listed clients include Claude Code, Cursor, Claude Desktop, Cline, and Codex. Installation targets macOS (Apple Silicon) via Homebrew and Linux (x86_64/ARM64) via a shell installer. Codebase indexing understands a dozen-plus languages through its parsing layer, enabling the server to present language-aware document slices to requesting agents.
How it fits into developer workflows and maintenance needs
The tool integrates into local developer workflows by detecting the active project directory automatically and applying project-level documentation policies defined in a policy.toml file. It also runs semantic linting to flag broken links and stale markers and supports multiple agents concurrently. Teams should expect an initial curation phase to align documentation and policies, since the quality of served context depends on how well docs are organized and maintained.
Suited to teams that commit to curated documentation and disciplined maintenance
The tool is a practical option for engineering teams that need controlled, on-demand context for coding agents, because it reduces prompt noise and narrows what agents retrieve. Expect an initial investment in documentation curation and routine upkeep to keep relevance high. A practical tip: pair narrow, explicit agent prompts with regular linting runs so the server serves concise, up-to-date fragments rather than large, noisy documents.
